Class D FAQ

[pma]

Yes

You also need to add a power stage that will distort not very nicely.
B. Putzeys toyed with the idea of a power DAC taking DSD as input and without feedback. He ended with something like this:
View attachment 135949
Source: A true One-Bit Power D/A Converter

A. Would be easy, LPF + power amplifier. You can make linear amp at least as good as the best class D.

I was rather thinking about something like B., fast switching at power level and filter at high level. Nowadays class D amps are limited in BW, quite miserably, even the Purify. To me, power amp must have BW > 100 kHz and I do not care what anyone thinks about.

FYI, I made my 1st PWM amp in 1982.

 

[orchardaudio]

This topic is dedicated to Frequently Asked Questions about class D amplifiers. If you have been directed here the reason is that it is easier to concentrate the frequent answers here instead of repeating them in several topics.


Q: Class D power ratings cannot be true. How can these small and light "toy amplifiers" generate all these watts??

A: The key word is efficiency. That means less power is wasted as heat, and as a consequence, the manufacturers don't have to attach heavy heatsinks to the amplifier and build a large chassis. But these watts and amperes are all true, measured with Audio Precision gear.


Q: Yeah, but the continuous power is a fraction of the rated power. They simply cannot play 20 kHz full scale sine waves continuously for long! Look at the datasheet, they even admit it! Their FTC power rating is a fraction of what they advertise. This is a fraud!

A: Music signal is dynamic, its RMS it about -12dB from full scale. Also it has much more energy in the bass area, 20 kHz is usually -40dB. In practice, playing loud music in difficult speakers, class D amplifiers run cool and for long periods. More info here.


Q: This high frequency switching noise must be audible! If not directly, it will mirror artifacts into the hearing range. Will I get headaches listening to class D amplifiers?

A: The switching noise is typically around 400-500 kHz. This is way above what your tweeters can reproduce, and even if they could, it is way higher than what humans (and dogs and bats) can hear, so no headaches. There are no high frequency artifacts showing up on measurements, just check any IMD FFT graphs.


Q: But will this high frequency switching noise burn the tweeters of my precious speakers?

A: No, it will not damage your tweeters. The switching noise is very low in level, from the amplifier idle up to full power. More info here.


Q: But class D amplifiers have phase shift in high frequencies. Does the treble really come later than the rest of the music? I don't want to distort my music.

A: No, class D amplifiers do not have phase shift. They have a constant time delay that, when plotted against the diminishing wave period, looks like a curve. But in fact it is a constant delay for all frequencies. Music arrives at you ears all together. More info here.


Q: I read that class D amplifiers are only suitable for the bass range, and mids and treble are better served with a class A/AB amplifier. Is it true that the highs sound "metallic" with class D amplifiers?

A: It may have been true some decades ago because old designs had more distortion on the mids and highs compared to the bass, specially when demanded with more power. But modern high end designs have similar low levels of distortion in all frequencies and in all loads, so that they are fine for bass, mids and treble. In fact, the high frequency IMD from Purifi Eigentakt is state of the art compared to any topology, see here.


Q: I read about GaN transistors and how they are the future. With this technology we can shift the switching frequency way up to megahertz and solve all the issues, right? I will never have to worry about output filters and high frequency distortion again!

A: While the technology is interesting and promising, we have yet to see an amplifier that has significantly better measurements than what we have with silicon today. More info here.


Q: Class D amplifier manufacturers are not true "manufacturers". All these "assemblers" do is mounting boards into boxes, and connect cables and connectors. This is not true designing like a traditional amplifier manufacturer.

A: While this is partially true for some complete amplifier modules, other modules need more parts to work (buffer boards and opamps, auxiliary power supplies, etc). Some manufacturers also add more features (auto-on, variable gain, subwoofer out with low pass) or even integrate other functions (DAC, streaming, DSP, room correction, etc). As for the "romantic" view of designing and building amplifiers in house as it has always been done, see here.


Q: I am never buying a class D amplifier! Ever! I hate them! They will have to take my class A/class AB/tubes from my dead hands!

A: That's fine, no one is forcing you to buy a class D amplifier. Just please stop spreading FUD.

I am going to reference this myself, great write-up.

 

[restorer-john]

To me, power amp must have BW > 100 kHz and I do not care what anyone thinks about.

I agree.

 

[Rottmannash]

I stumbled upon this list https://docs.google.com/spreadsheets/d/1ZlTOYxmPs938gqHjtDABkWS-MApu7uJjzIGnJ2Elm6Y/edit#gid=0 looking at the signature of Matias and it just made me order two Hypex NC400 Mono. It will be my last try with Class D. If those things don't live up to the hype I'm going to stick to Class AB for eternity.

Why didn't you buy a Purifi amp if you wanted the best class D amp? Glad you found the monoblocs you like tho.

 

[anphex]

Why didn't you buy a Purifi amp if you wanted the best class D amp? Glad you found the monoblocs you like tho.

I'd have chosen Purifi if they would have had a nice DIY kit like hypex does. I know you can buy the parts to build a monoblock separately but it was to deterrend for me. This and also the specs of the NC400 and EVAL2 look about the same while the price is higher for the purifiy. The monoblocks were 1200 €. One EVAL2 Monoblock would be about 900 € alone.
And if I wanted to go "ultimate" I'd go with an AHB2.

So the Hypex is the price/value winner for me.
Still, I'd love to try out a pure Purify build one day.

 

[voodooless]

To me, power amp must have BW > 100 kHz and I do not care what anyone thinks about.

I agree.

Doesn't really matter what anyone thinks, the question is: Why?

 

[Koeitje]

A. Would be easy, LPF + power amplifier. You can make linear amp at least as good as the best class D.

I was rather thinking about something like B., fast switching at power level and filter at high level. Nowadays class D amps are limited in BW, quite miserably, even the Purify. To me, power amp must have BW > 100 kHz and I do not care what anyone thinks about.

FYI, I made my 1st PWM amp in 1982.

Do you play music for your cats?

 

[restorer-john]

Doesn't really matter what anyone thinks, the question is: Why?

Waveform fidelity. The entire premise and reason for existence of high fidelity audio.

You cannot describe the perfect reproduction of an arbitrary waveform within the audible spectrum without a bandwidth that exceeds that spectrum by at least an order of magnitude (x10).

My analyser has a 200kHz measurement bandwidth. Why? So it can accurately catch all the distortion harmonics of a 20kHz waveform.

 

[restorer-john]

Do you play music for your cats?

What a silly comment. You can do way better. I know you can.

 

[DonH56]

Reasons for extended bandwidth include in-band flatness (e.g. 0.1 dB to 20 kHz requires ~131 kHz bandwidth in a system with first-order roll-off), lower phase change (shift) in the audio band, enough slew rate to prevent transient distortion and large-signal steady-state distortion, and perhaps most importantly sufficient bandwidth to provide the gain for feedback to reduce the distortion to negligible levels. Feedback requires internal gain to reduce the distortion, and providing that gain means you must have much greater bandwidth than the highest signal frequency you are attempting to compensate. For modest fidelity an octave (factor of two) may be enough, but high-linearity designs often design for a factor of ten greater bandwidth than the signal itself.

FWIWFM - Don

Edit: @restorer-john beat me to it.

Edit 2: I was skimming new posts and missed this was about class D amplifiers. The rules are a little different for that topology, though for internal feedback loops some of the comments are still relevant. Thanks @boXem | audio for catching that blunder.

 

[oivavoi]

Reasons for extended bandwidth include in-band flatness (e.g. 0.1 dB to 20 kHz requires ~131 kHz bandwidth in a system with first-order roll-off), lower phase change (shift) in the audio band, enough slew rate to prevent transient distortion and large-signal steady-state distortion, and perhaps most importantly sufficient bandwidth to provide the gain for feedback to reduce the distortion to negligible levels. Feedback requires internal gain to reduce the distortion, and providing that gain means you must have much greater bandwidth than the highest signal frequency you are attempting to compensate. For modest fidelity an octave (factor of two) may be enough, but high-linearity designs often design for a factor of ten greater bandwidth than the signal itself.

FWIWFM - Don

Edit: @restorer-john beat me to it.

Interesting. It's still a question of audibility though. I would be really interested in seeing controlled tests where people are able to hear a difference between transducers being driven by any of the good class D amps and an AB amp with higher bandwidth. I know that @pma was able to identify the Ayima amplifier in a blind test, but how about the offerings from hypex, pascal or yamaha?

 

[voodooless]

Waveform fidelity. The entire premise and reason for existence of high fidelity audio.

You cannot describe the perfect reproduction of an arbitrary waveform within the audible spectrum without a bandwidth that exceeds that spectrum by at least an order of magnitude (x10).

So Purify:

The HDA2 has the same HF rise, just an order of magnitude lower due to better distortion and noise figures. It is 1 dB down at 100 kHz, so would probably qualify as "good" bandwidth by your book.

The profile looks quite similar. Obviously, there are also AB amps with a flat THD vs frequency plot like @pma amps in his signature, but on the other hand, we have a lot of AB amps tested that perform not as good as the nCore or Purifies in this regard. And also we've seen a lot of class D amps that perform much worse as well.

Reasons for extended bandwidth include in-band flatness (e.g. 0.1 dB to 20 kHz requires ~131 kHz bandwidth in a system with first-order roll-off)

Class D amps usually don't have 1st order roll-off, do they?

, lower phase change (shift) in the audio band,

Yes, that is indeed a thing even if probably not audible.

enough slew rate to prevent transient distortion and large-signal steady-state distortion,

Is that a thing with class D amps? To quote Bruno:

Slew rate in class D amplifiers is completely unrelated to that found in linear amplifiers

The "slew rate" is limited by the output filter. Apply a voltage step from -Vcc to +Vcc to the output filter and look at the output. The bandwidth limiting action of the filter necessarily limits the dV/dt found there. That doesn't mean that the filter or anything else has gone non-linear. It's just that the bandwidth is limited, nothing more, nothing less. Slew rate in a class D amplifier is a completely linear effect. Here, power 6 bandwidth equals small signal bandwidth by definition, because there is no non-linear effect involved.

and perhaps most importantly sufficient bandwidth to provide the gain for feedback to reduce the distortion to negligible levels. Feedback requires internal gain to reduce the distortion, and providing that gain means you must have much greater bandwidth than the highest signal frequency you are attempting to compensate.

I think we've seen a lot of examples that show very low distortion figures. Sure, HF is usually a bit worse...

There is always a compromise.. question is: which ones do you want to live with?

 

[Matias]

There is always a compromise.. question is: which ones do you want to live with?

Exactly. Bruno's compromise of maximizing in band performance before the ultrasonic spectrum is an intelligent decision, more sensible than 200 kHz extension and harmonics... At the end of the day an amp is for listening to music, not for measuring irrelevant signals.

 

[antcollinet]

Doesn't really matter what anyone thinks, the question is: Why?

I was going to ask that also.

 

[boXem]

Reasons for extended bandwidth include in-band flatness (e.g. 0.1 dB to 20 kHz requires ~131 kHz bandwidth in a system with first-order roll-off), lower phase change (shift) in the audio band, enough slew rate to prevent transient distortion and large-signal steady-state distortion, and perhaps most importantly sufficient bandwidth to provide the gain for feedback to reduce the distortion to negligible levels. Feedback requires internal gain to reduce the distortion, and providing that gain means you must have much greater bandwidth than the highest signal frequency you are attempting to compensate. For modest fidelity an octave (factor of two) may be enough, but high-linearity designs often design for a factor of ten greater bandwidth than the signal itself.

FWIWFM - Don

Edit: @restorer-john beat me to it.

Completely agree. For classical amps with first order loops.
Self oscillating amplifiers are a completely different paradigm. These use high order loops and their external behavior is only linked to the difference between the number of poles and zeros from the loop.
Ncore have a first order behavior, Eigentakt a second order behavior.
The Eigentakt 75 dB loop gain at 20 kHz gets Purifi advertising their modules as having an equivalent GBW product of 110 MHz.

 

[rdenney]

It seems to me that designing for ten times the audio band is a design rule of thumb to account for a range of things that affect performance downstream. The NC502MP claims a 0,-3 dB bandwidth of 50 KHz, but does roll off frequency response in the audio band by more than 0.1 dB near 20 KHz. So, the “limited” bandwidth means it’s 0.6 dB down at 20 KHz. (I put “limited” in quotes because I doubt the audibility of this roll off.)

But the graph is post-filtration, right?

Note the vertical scale. Plot this on a graph with 5-dB divisions and the frequency response would look flatter over a higher range.

Rick “wondering whether power bandwidth includes filtration or not” Denney

 

[DonH56]

Completely agree. For classical amps with first order loops.
Self oscillating amplifiers are a completely different paradigm. These use high order loops and their external behavior is only linked to the difference between the number of poles and zeros from the loop.
Ncore have a first order behavior, Eigentakt a second order behavior.
The Eigentakt 75 dB loop gain at 20 kHz gets Purifi advertising their modules as having an equivalent GBW product of 110 MHz.

Yah, oops. Skimming new posts and missed this was a thread on class D amplifiers. Added an edit to my post.

 

[boXem]

It seems to me that designing for ten times the audio band is a design rule of thumb to account for a range of things that affect performance downstream. The NC502MP claims a 0,-3 dB bandwidth of 50 KHz, but does roll off frequency response in the audio band by more than 0.1 dB near 20 KHz. So, the “limited” bandwidth means it’s 0.6 dB down at 20 KHz. (I put “limited” in quotes because I doubt the audibility of this roll off.)

But the graph is post-filtration, right?

Note the vertical scale. Plot this on a graph with 5-dB divisions and the frequency response would look flatter over a higher range.

Rick “wondering whether power bandwidth includes filtration or not” Denney

Graph is post filtration.

 

[EJ3]

Tbh people who intentionally buy Class A or Class A/B amps in 2021 are the anti-maskers of the audio world.

They measure worse (at the same price), they cost more (for the same wattage) and they consume more energy.

I would like to know the cradle to cradle cost (how many are just disposed because people can't fix them or it is just cheaper to buy another & throw the old one into a landfill in some other country. What is the true cost in wasted time (that you cannot get back, stereo down time, time away from your family or normal activities because you have to dispose the CLASS D when you could repair a class AB (G or H) that has been around for three decades. I don't thing the minor efficiencies over class A/B/G/H make up for the overall cost when you get to 3 generations of audio equipment that did not have to be disposed of (landfills, perhaps across the ocean, in the land of other countries). Consider the people there & their environment..

 

[pma]

Tbh people who intentionally buy Class A or Class A/B amps in 2021 are the anti-maskers of the audio world.

The comparison fails. I wear masks ewerywhere where it is requested and got fully vaccinated in May